An assemble pattern relevance feedback technique for an automatic image annotation

In image retrieval system the basic method is the content based image retrieval. Different from traditional database queries. Content based multimedia retrieval queries are vague, that creates complex for the users to show their exact information to be made available in comfort and right query. By inducing relevance feedback techniques in content based image retrieval, more clear results can be achieved from the account of user's feedback. However the previous relevance feedback based content based image retrieval are refined in retrieval results particularly in large-scale image database. It becomes inefficient and not applicable in real world applications. Assemble Pattern Relevance Feedback (APRF) is implemented to achieve more efficiency, high performance, effectiveness, less expensive in time and memory, reduced generalization error of classification models. These are all improved by using assemble pattern invented from user query log. The three kinds of query breeding scheme are Query Point Position (QPP), Query Reconcilation (QC), Query Stretching (QS) is to make the search space towards user's animus. By using APRF method, high quality of image retrieval on relevance feedback can be attained in minimum number of feedbacks.